1 /* Copyright (c) 2007-2023. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
6 #include "src/mc/remote/CheckerSide.hpp"
7 #include "src/mc/explo/Exploration.hpp"
8 #include "src/mc/explo/LivenessChecker.hpp"
9 #include "src/mc/sosp/RemoteProcessMemory.hpp"
10 #include "xbt/system_error.hpp"
13 #include <sys/personality.h>
14 #include <sys/prctl.h>
17 #include <boost/tokenizer.hpp>
20 #include <sys/ptrace.h>
24 #define WAITPID_CHECKED_FLAGS __WALL
26 #define WAITPID_CHECKED_FLAGS 0
29 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkerside, mc, "MC communication with the application");
31 static simgrid::config::Flag<std::string> _sg_mc_setenv{
32 "model-check/setenv", "Extra environment variables to pass to the child process (ex: 'AZE=aze;QWE=qwe').", "",
33 [](std::string_view value) {
34 xbt_assert(value.empty() || value.find('=', 0) != std::string_view::npos,
35 "The 'model-check/setenv' parameter must be like 'AZE=aze', but it does not contain an equal sign.");
38 namespace simgrid::mc {
40 XBT_ATTRIB_NORETURN static void run_child_process(int socket, const std::vector<char*>& args, bool need_ptrace)
42 /* On startup, simix_global_init() calls simgrid::mc::Client::initialize(), which checks whether the MC_ENV_SOCKET_FD
43 * env variable is set. If so, MC mode is assumed, and the client is setup from its side
47 // Make sure we do not outlive our parent
50 xbt_assert(sigprocmask(SIG_SETMASK, &mask, nullptr) >= 0, "Could not unblock signals");
51 xbt_assert(prctl(PR_SET_PDEATHSIG, SIGHUP) == 0, "Could not PR_SET_PDEATHSIG");
53 // Make sure that the application process layout is not randomized, so that the info we gather is stable over re-execs
54 if (personality(ADDR_NO_RANDOMIZE) == -1) {
55 XBT_ERROR("Could not set the NO_RANDOMIZE personality");
56 throw xbt::errno_error();
60 // Remove CLOEXEC to pass the socket to the application
61 int fdflags = fcntl(socket, F_GETFD, 0);
62 xbt_assert(fdflags != -1 && fcntl(socket, F_SETFD, fdflags & ~FD_CLOEXEC) != -1,
63 "Could not remove CLOEXEC for socket");
65 setenv(MC_ENV_SOCKET_FD, std::to_string(socket).c_str(), 1);
67 setenv("MC_NEED_PTRACE", "1", 1);
69 /* Setup the tokenizer that parses the cfg:model-check/setenv parameter */
70 using Tokenizer = boost::tokenizer<boost::char_separator<char>>;
71 boost::char_separator<char> semicol_sep(";");
72 boost::char_separator<char> equal_sep("=");
73 Tokenizer token_vars(_sg_mc_setenv.get(), semicol_sep); /* Iterate over all FOO=foo parts */
74 for (const auto& token : token_vars) {
75 std::vector<std::string> kv;
76 Tokenizer token_kv(token, equal_sep);
77 for (const auto& t : token_kv) /* Iterate over 'FOO' and then 'foo' in that 'FOO=foo' */
79 xbt_assert(kv.size() == 2, "Parse error on 'model-check/setenv' value %s. Does it contain an equal sign?",
81 XBT_INFO("setenv '%s'='%s'", kv[0].c_str(), kv[1].c_str());
82 setenv(kv[0].c_str(), kv[1].c_str(), 1);
85 /* And now, exec the child process */
87 while (args[i] != nullptr && args[i][0] == '-')
90 xbt_assert(args[i] != nullptr,
91 "Unable to find a binary to exec on the command line. Did you only pass config flags?");
93 execvp(args[i], args.data() + i);
94 XBT_CRITICAL("The model-checked process failed to exec(%s): %s.\n"
95 " Make sure that your binary exists on disk and is executable.",
96 args[i], strerror(errno));
97 if (strchr(args[i], '=') != nullptr)
98 XBT_CRITICAL("If you want to pass environment variables to the application, please use --cfg=model-check/setenv:%s",
101 xbt_die("Aborting now.");
104 static void wait_application_process(pid_t pid)
106 XBT_DEBUG("Waiting for the model-checked process");
109 // The model-checked process SIGSTOP itself to signal it's ready:
110 xbt_assert(waitpid(pid, &status, WAITPID_CHECKED_FLAGS) == pid && WIFSTOPPED(status) && WSTOPSIG(status) == SIGSTOP,
111 "Could not wait model-checked process");
115 ptrace(PTRACE_SETOPTIONS, pid, nullptr, PTRACE_O_TRACEEXIT);
116 ptrace(PTRACE_CONT, pid, 0, 0);
118 ptrace(PT_CONTINUE, pid, (caddr_t)1, 0);
120 #error "no ptrace equivalent coded for this platform"
122 xbt_assert(errno == 0,
123 "Ptrace does not seem to be usable in your setup (errno: %d). "
124 "If you run from within a docker, adding `--cap-add SYS_PTRACE` to the docker line may help. "
125 "If it does not help, please report this bug.",
129 void CheckerSide::setup_events()
131 if (base_ != nullptr)
132 event_base_free(base_.get());
133 auto* base = event_base_new();
136 socket_event_ = event_new(
137 base, get_channel().get_socket(), EV_READ | EV_PERSIST,
138 [](evutil_socket_t, short events, void* arg) {
139 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
140 if (events == EV_READ) {
141 std::array<char, MC_MESSAGE_LENGTH> buffer;
142 ssize_t size = recv(checker->get_channel().get_socket(), buffer.data(), buffer.size(), MSG_DONTWAIT);
144 XBT_ERROR("Channel::receive failure: %s", strerror(errno));
146 throw simgrid::xbt::errno_error();
149 if (size == 0) // The app closed the socket. It must be dead by now.
150 checker->handle_waitpid();
151 else if (not checker->handle_message(buffer.data(), size))
152 checker->break_loop();
154 xbt_die("Unexpected event");
158 event_add(socket_event_, nullptr);
160 signal_event_ = event_new(
161 base, SIGCHLD, EV_SIGNAL | EV_PERSIST,
162 [](evutil_socket_t sig, short events, void* arg) {
163 auto checker = static_cast<simgrid::mc::CheckerSide*>(arg);
164 if (events == EV_SIGNAL) {
166 checker->handle_waitpid();
168 xbt_die("Unexpected signal: %d", sig);
170 xbt_die("Unexpected event");
174 event_add(signal_event_, nullptr);
177 CheckerSide::CheckerSide(const std::vector<char*>& args, bool need_memory_info) : running_(true)
179 bool need_ptrace = not need_memory_info;
181 // Create an AF_LOCAL socketpair used for exchanging messages between the model-checker process (ancestor)
182 // and the application process (child)
184 xbt_assert(socketpair(AF_LOCAL, SOCK_SEQPACKET | SOCK_CLOEXEC, 0, sockets) != -1, "Could not create socketpair");
187 xbt_assert(pid_ >= 0, "Could not fork model-checked process");
189 if (pid_ == 0) { // Child
191 run_child_process(sockets[0], args, need_ptrace);
195 // Parent (model-checker):
197 channel_.reset_socket(sockets[1]);
201 wait_application_process(pid_);
203 // Request the initial memory on need
204 if (need_memory_info) {
205 channel_.send(MessageType::INITIAL_ADDRESSES);
206 s_mc_message_initial_addresses_reply_t answer;
207 ssize_t answer_size = channel_.receive(answer);
208 xbt_assert(answer_size != -1, "Could not receive message");
209 xbt_assert(answer.type == MessageType::INITIAL_ADDRESSES_REPLY,
210 "The received message is not the INITIAL_ADDRESS_REPLY I was expecting but of type %s",
211 to_c_str(answer.type));
212 xbt_assert(answer_size == sizeof answer, "Broken message (size=%zd; expected %zu)", answer_size, sizeof answer);
214 /* We now have enough info to create the memory address space */
215 remote_memory_ = std::make_unique<simgrid::mc::RemoteProcessMemory>(pid_, answer.mmalloc_default_mdp);
221 CheckerSide::~CheckerSide()
223 event_del(socket_event_);
224 event_free(socket_event_);
225 event_del(signal_event_);
226 event_free(signal_event_);
230 xbt_assert(kill(get_pid(), SIGKILL) == 0);
233 waitpid(get_pid(), nullptr, 0);
234 xbt_assert(errno == 0);
235 } while (EAGAIN == errno);
239 void CheckerSide::finalize(bool terminate_asap)
241 s_mc_message_int_t m = {};
242 m.type = MessageType::FINALIZE;
243 m.value = terminate_asap;
244 xbt_assert(get_channel().send(m) == 0, "Could not ask the app to finalize on need");
246 s_mc_message_t answer;
247 ssize_t s = get_channel().receive(answer);
248 xbt_assert(s != -1, "Could not receive answer to FINALIZE");
249 xbt_assert(s == sizeof answer, "Broken message (size=%zd; expected %zu)", s, sizeof answer);
250 xbt_assert(answer.type == MessageType::FINALIZE_REPLY,
251 "Received unexpected message %s (%i); expected MessageType::FINALIZE_REPLY (%i)", to_c_str(answer.type),
252 (int)answer.type, (int)MessageType::FINALIZE_REPLY);
255 void CheckerSide::dispatch_events() const
257 event_base_dispatch(base_.get());
260 void CheckerSide::break_loop() const
262 event_base_loopbreak(base_.get());
265 bool CheckerSide::handle_message(const char* buffer, ssize_t size)
267 s_mc_message_t base_message;
268 xbt_assert(size >= (ssize_t)sizeof(base_message), "Broken message. Got only %ld bytes.", size);
269 memcpy(&base_message, buffer, sizeof(base_message));
271 switch (base_message.type) {
272 case MessageType::IGNORE_HEAP: {
273 if (remote_memory_ != nullptr) {
274 s_mc_message_ignore_heap_t message;
275 xbt_assert(size == sizeof(message), "Broken message");
276 memcpy(&message, buffer, sizeof(message));
278 IgnoredHeapRegion region;
279 region.block = message.block;
280 region.fragment = message.fragment;
281 region.address = message.address;
282 region.size = message.size;
283 get_remote_memory()->ignore_heap(region);
285 XBT_INFO("Ignoring a IGNORE_HEAP message because we don't need to introspect memory.");
290 case MessageType::UNIGNORE_HEAP: {
291 if (remote_memory_ != nullptr) {
292 s_mc_message_ignore_memory_t message;
293 xbt_assert(size == sizeof(message), "Broken message");
294 memcpy(&message, buffer, sizeof(message));
295 get_remote_memory()->unignore_heap((void*)message.addr, message.size);
297 XBT_INFO("Ignoring an UNIGNORE_HEAP message because we don't need to introspect memory.");
302 case MessageType::IGNORE_MEMORY: {
303 if (remote_memory_ != nullptr) {
304 s_mc_message_ignore_memory_t message;
305 xbt_assert(size == sizeof(message), "Broken message");
306 memcpy(&message, buffer, sizeof(message));
307 get_remote_memory()->ignore_region(message.addr, message.size);
309 XBT_INFO("Ignoring an IGNORE_MEMORY message because we don't need to introspect memory.");
314 case MessageType::STACK_REGION: {
315 if (remote_memory_ != nullptr) {
316 s_mc_message_stack_region_t message;
317 xbt_assert(size == sizeof(message), "Broken message");
318 memcpy(&message, buffer, sizeof(message));
319 get_remote_memory()->stack_areas().push_back(message.stack_region);
321 XBT_INFO("Ignoring an STACK_REGION message because we don't need to introspect memory.");
326 case MessageType::REGISTER_SYMBOL: {
327 s_mc_message_register_symbol_t message;
328 xbt_assert(size == sizeof(message), "Broken message");
329 memcpy(&message, buffer, sizeof(message));
330 xbt_assert(not message.callback, "Support for client-side function proposition is not implemented.");
331 XBT_DEBUG("Received symbol: %s", message.name.data());
333 LivenessChecker::automaton_register_symbol(*get_remote_memory(), message.name.data(), remote((int*)message.data));
337 case MessageType::WAITING:
340 case MessageType::ASSERTION_FAILED:
341 Exploration::get_instance()->report_assertion_failure();
345 xbt_die("Unexpected message from model-checked application");
350 void CheckerSide::wait_for_requests()
352 /* Resume the application */
353 if (get_channel().send(MessageType::CONTINUE) != 0)
354 throw xbt::errno_error();
355 clear_memory_cache();
361 void CheckerSide::clear_memory_cache()
364 remote_memory_->clear_cache();
367 void CheckerSide::handle_waitpid()
369 XBT_DEBUG("Check for wait event");
372 while ((pid = waitpid(-1, &status, WNOHANG)) != 0) {
374 if (errno == ECHILD) { // No more children:
375 xbt_assert(not this->running(), "Inconsistent state");
378 xbt_die("Could not wait for pid: %s", strerror(errno));
382 if (pid == get_pid()) {
383 // From PTRACE_O_TRACEEXIT:
385 if (status >> 8 == (SIGTRAP | (PTRACE_EVENT_EXIT << 8))) {
386 unsigned long eventmsg;
387 xbt_assert(ptrace(PTRACE_GETEVENTMSG, pid, 0, &eventmsg) != -1, "Could not get exit status");
388 status = static_cast<int>(eventmsg);
389 if (WIFSIGNALED(status)) {
391 Exploration::get_instance()->report_crash(status);
396 // We don't care about non-lethal signals, just reinject them:
397 if (WIFSTOPPED(status)) {
398 XBT_DEBUG("Stopped with signal %i", (int)WSTOPSIG(status));
401 ptrace(PTRACE_CONT, pid, 0, WSTOPSIG(status));
403 ptrace(PT_CONTINUE, pid, (caddr_t)1, WSTOPSIG(status));
405 xbt_assert(errno == 0, "Could not PTRACE_CONT");
408 else if (WIFSIGNALED(status)) {
410 Exploration::get_instance()->report_crash(status);
411 } else if (WIFEXITED(status)) {
412 XBT_DEBUG("Child process is over");
419 } // namespace simgrid::mc